88Z AI, 26 June — DeepReinforce has released the open-source Agentic programming model Ornith-1.0, built on Gemma 4 and Qwen 3.5 bases with four sizes ranging from 9B to 397B. The series performs strongly on Terminal-Bench 2.1 and SWE-Bench Verified benchmarks, with the flagship 397B version surpassing Claude Opus 4.7 and the 35B MoE version outperforming Qwen 3.5-397B.
At a glance
- Ornith-1.0 is built on Gemma 4 and Qwen 3.5 bases, offering four sizes from 9B to 397B
- The flagship 397B version surpasses Claude Opus 4.7, while the 35B MoE edition outperforms Qwen 3.5-397B
The update
The DeepReinforce team has officially launched Ornith-1.0, an open-source series of Agentic programming models. Built on the Gemma 4 and Qwen 3.5 foundations, the model utilizes a self-improving training framework for optimization. Ornith-1.0 is available in four specifications ranging from 9B to 397B, designed to meet diverse development needs. In terms of performance evaluation, the series achieved state-of-the-art results among peers on the Terminal-Bench 2.1 and SWE-Bench Verified benchmarks. The flagship 397B version scored 77.5 and 82.4 respectively, surpassing the performance of Claude Opus 4.7. Furthermore, the 35B MoE version achieved a score of 64.4, outperforming the larger Qwen 3.5-397B model, demonstrating significant efficiency advantages.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- Ornith-1.0Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.